A patient leaves a consultation with a prescription and a diagnosis. On the way back, the questions arrive, those he didn’t dare to ask, those he forgot…
or even those to which the doctor did not have time to answer. For a long time, he only had two options, call back a crowded office or get lost in anxiety-inducing forums. A third way has emerged in just a few months, and it challenges an obvious fact that medicine did not like to face, namely that medical time is no longer enough to cover the need for patient support.
The silence that no one had filled
More than a third of French people have already questioned generative artificial intelligence on a health subject, a proportion which rises to 68% among 18-24 year olds (FLASHS survey for Galeon – March 2025). This figure does not measure a distrust of doctors, but rather the absence of a contact between two appointments, when patients need it most.
A patient undergoing chemotherapy wonders if his fatigue is normal on a Tuesday evening. A woman in the process of assisted reproduction wonders about her protocol at 11 p.m. A diabetic patient doesn’t know what to cook with his new treatment. These questions are legitimate and urgent, but they have no structural addressee, since no doctor can give their personal number without being overwhelmed the next day. The patient who questions an AI at 11 p.m. is therefore not looking for a substitute for the doctor, he is simply looking for someone who will answer him.
What AI really does, and what it doesn’t do
The most recent language models display performance that, on paper, is impressive. Researchers (Harvard Medical School and Beth Israel Deaconess Medical Center, April 2026) evaluated OpenAI’s o1 model on hundreds of typical clinical cases. The correct diagnosis was among the AI’s hypotheses in 78% of cases, and its clinical reasoning was judged optimal in 98% of cases, compared to only 35% for doctors to whom the same cases were submitted. But reducing the contribution of AI to diagnosis would mean missing the point.
Its true value lies elsewhere, in everything that medicine has never had the time to do correctly, such as explaining a treatment with words adapted to the listener, deciphering a photographed prescription in a few seconds with its interactions and its expected effects, or even accompanying the anguish of a chronic patient at three in the morning that no one else is listening to at that time. It is not a question of replacement medicine but of attention made available, free of charge, at the precise moment when it was lacking.
The real shift, from distrust to method
The paradox is that it is now also doctors who use these tools, silently at first, then less and less, to check a drug interaction, refresh their memory on a rare protocol or prepare an explanation for a worried patient. This systematic verification is in no way a professional error and on the contrary constitutes an additional guarantee of rigor.
The received idea to abandon is therefore not that AI would threaten the healthcare relationship, but rather the idea that medical time, as it is organized today, would still be sufficient for this relationship. Administrative tasks take up between 22 and 33% of a general practitioner’s daily time (DUMAS thesis, 2024, “Impact of administrative work during general medicine consultations”), and each minute spent on paperwork is a minute spent listening and examining. Medical trust does not survive opacity; it is built beforehand and never afterwards.
Pharmaceutical laboratories, mutual insurance companies and health platforms can no longer treat this use as a passing curiosity, especially since certain appointment scheduling platforms are already starting to exploit AI to anticipate appointments at risk of absence, with encouraging initial results. The question is therefore no longer whether artificial intelligence will transform the relationship between patients and the health system, since it already does so every evening, in millions of discreet conversations.
It remains to be seen who will carry out this transformation with the demands it deserves, by training healthcare teams, establishing legitimate uses and building trust rather than enduring it. This project is no less strategic than any other priority in the health sector, and a medicine that could rely on these tools without ever delegating its responsibility to them would not be colder medicine but available medicine.
